English

Compositional Transfer in Hierarchical Reinforcement Learning

Machine Learning 2020-05-20 v3 Artificial Intelligence Robotics Machine Learning

Abstract

The successful application of general reinforcement learning algorithms to real-world robotics applications is often limited by their high data requirements. We introduce Regularized Hierarchical Policy Optimization (RHPO) to improve data-efficiency for domains with multiple dominant tasks and ultimately reduce required platform time. To this end, we employ compositional inductive biases on multiple levels and corresponding mechanisms for sharing off-policy transition data across low-level controllers and tasks as well as scheduling of tasks. The presented algorithm enables stable and fast learning for complex, real-world domains in the parallel multitask and sequential transfer case. We show that the investigated types of hierarchy enable positive transfer while partially mitigating negative interference and evaluate the benefits of additional incentives for efficient, compositional task solutions in single task domains. Finally, we demonstrate substantial data-efficiency and final performance gains over competitive baselines in a week-long, physical robot stacking experiment.

Keywords

Cite

@article{arxiv.1906.11228,
  title  = {Compositional Transfer in Hierarchical Reinforcement Learning},
  author = {Markus Wulfmeier and Abbas Abdolmaleki and Roland Hafner and Jost Tobias Springenberg and Michael Neunert and Tim Hertweck and Thomas Lampe and Noah Siegel and Nicolas Heess and Martin Riedmiller},
  journal= {arXiv preprint arXiv:1906.11228},
  year   = {2020}
}

Comments

Robotics Science and Systems 2020

R2 v1 2026-06-23T10:04:32.190Z